For results that have...

But don't include...

Abstract

Click to add/remove this article to your list of 'My Favorites' A Clinical Study Of Meal Size Estimation Analysis Using Continuous Glucose Monitoring (CGM) Data

Year: 2008

Abstract Number: 1965-PO

Authors: HYUNJIN LEE, KIMBERLY CASWELL, EYAL DASSAU, FRANCIS J. DOYLE III, BRUCE A. BUCKINGHAM, B. W. BEQUETTE, Troy, NY, Stanford, CA, Santa Barbara, CA

Institutions: Santa Barbara, CA; Stanford, CA; Troy, NY

Results: An effective structure for a closed-loop artificial pancreas is a hybrid of feedback to glucose and feed-forward to meals as a combined control action. The feed-forward portion is particularly challenging to implement when meal sizes are unknown and requires detection of a meal event and an estimate of the size of the meal.
We have developed a meal detection and meal size estimation algorithm based on optimal estimation theory and discrete-time signal processing methods. A set of threshold values for the first and second derivatives of glucose with respect to time are found to detect a meal within 15-45 minutes of actual meal consumption. When a meal is detected, a finite impulse response filter, based on the estimated second derivative of glucose with time, is used to estimate the meal size (grams of carbohydrate).
We evaluated the performance of this algorithm utilizing four ambulatory subjects with type 1 diabetes (17 to 38 years old) wearing a Navigator and recording the carbohydrate content of a meal while withholding insulin delivery for 30 minutes after meal consumption. Seventeen meals with a mean meal size of 59 grams of CHO (range 14 to 110 grams) were used for this study. The meal size estimation filter was trained using 4 meals (the median detection time was 34 minutes) from 2 individuals. The algorithm was validated against the remaining 13 meals: 3 meals were not detected (15%), the median (25%tile, 75% tile) detection time was 24 (17, 39) minutes, and the median error in estimating the meal size was -15% (-30%, 38%) with a mean error of 6%. Using an in silico model, simulations revealed a significant improvement in the closed-loop performance of a model predictive control algorithm, and postprandial mean glucose levels were reduced from 217 to 145 mg/dL and the maximum postprandial glucose was reduced from 333 to 232 mg/dL.
Future clinical studies will be needed to test the performance of this algorithm in a clinical setting using an “artificial pancreas.”